Hi folks.

Some of you might have seen my 3D Mandelbrot still / animation on the
binaries lists a while back... I set the thing running one night at 640 x
480, and came back the next morning. After 11 hours of rendering, it had
managed 96 frames. But then I added a drop-dead simple optimisation: instead
of one box {} for each "pixel", I arranged that where a row of pixels are
all the same height, the macro groups them all into a single box {} object.

The result?

Well, the time taken to generate the fractal include file didn't change
much. But the include file went from being just over 10 MB down to a tiny
860 KB!!! POV-Ray's scene parse was delighted. The object count went from
millions to thousands.

Actually, although I ment to coalesce (is that how you spell it?!?) ROWS of
pixels together, by mistake I actually did COLUMNS instead. However, from
the angle the camera looks at it, that actually means that the boxes are
short-edge on to the camera, so the vista buffer was delighted too! Result?
I finished the entire 400 frame animation in less than an hour! (And that
with aggressive antialiasing settings.)

Feeling pretty chuffed here...

Still fiddling with lighting and camera angle and stuff, will hopefully post
my results. (Oh yeah, and I really must make the lights FADE out at the end,
not just stop...)

Anyway, just a random thought... I remember somewhere someone saying that
Skyvase isn't a very meaningful benchmark render any more, cos it's too
fast! LOL. IIRC, the benchmark scene has been changed now anyway, but I was
wondering... I wonder if my fractal scene would make a good test render? I
think it would, for several reasons:

First, it's slow ;-)

But more than that, it doesn't just test RENDER speed... The bit where it
calculates the fractal tests how fast POV-Ray can execute SDL macros. The
bit where it loads the fractal data in tests how fast the scene parser is.
And yes, the bit where it renders tests how well the vista and light buffers
work.

An a vaguely-related note, is that why reflections are so slow to calculate?
Because they can't use the vista buffer to reduce the number of ray
intersection tests?

Anyway, I think I've wondered wildly off topic quite enough so far... I'm
going to shut up for a while now ;-)

Andrew.

Post a reply to this message

Andrew Coppin <orp### [at] btinternetcom> wrote:
> I wonder if my fractal scene would make a good test render?

  If it uses only boxes, I don't think it's diversified enough.

  The idea in the current official benchmark scene is that it uses many
of the most commonly used primitives and features of POV-Ray. Your scene
would probably simply test the rendering speed of boxes.

-- 
#macro N(D)#if(D>99)cylinder{M()#local D=div(D,104);M().5,2pigment{rgb M()}}
N(D)#end#end#macro M()<mod(D,13)-6mod(div(D,13)8)-3,10>#end blob{
N(11117333955)N(4254934330)N(3900569407)N(7382340)N(3358)N(970)}//  - Warp -

Post a reply to this message

>   If it uses only boxes, I don't think it's diversified enough.
>
>   The idea in the current official benchmark scene is that it uses many
> of the most commonly used primitives and features of POV-Ray. Your scene
> would probably simply test the rendering speed of boxes.

You have a point... tests raw speed pretty well, but doesn't test POV-Rays'
various features very much.

Thanx for that thought.
Andrew.

Post a reply to this message